Measurement error of state variables creates substantial bias in results of demographic population models.

Abstract

Integral projection and matrix population models are commonly used in ecological and conservation studies to assess the health and extinction risk of populations. These models use one (or more) measurable state variable(s), such as size or age, to predict individual performance, which, ideally, is the sole determinant of an individual's expected fate. However, even if ecologists successfully identify and measure the observable state variable(s) that best predicts individual fate, we are rarely, if ever, able to perfectly measure state for many species, especially those with size structure, where total plant biomass or starch stores, for example, may be the best predictors of fate. Here, we used a series of simulations to test how this imperfect quantification of actual state ("measurement error") leads to inaccurate prediction of state-dependent fates and influences the predictions of structured population models. We simulated 10yr of best practice field data collection using known vital rate functions and incorporated measurement error of different magnitudes and types (completely random, temporal, and individual based) for two size-structured life histories. We found that even for conservative error rates, most types of measurement error increased the median predicted population growth rate by 1-2% growth per year. However, the magnitude of this error differed substantially with life history strategy and error type, with some scenarios resulting in >8% median overestimation of population growth rate. This effect arises largely from the well-known econometrics problem of "regression dilution" (overestimation of the intercept and underestimation of the slope of a regression when the predictor variable is measured with error), which in our simulations typically results in overly optimistic predictions of small or young individuals' vital rates. Our results suggest that the problem of measurement error for state variables, present in many demographic studies but virtually unacknowledged in the ecological literature, may lead to substantial misestimation of population behavior, resulting in erroneous inferences about not only growth, but also extinction risk and other aspects of population dynamics.